philippon reshef skill biased development w13437

7/31/2019 Philippon Reshef Skill Biased Development w13437

1/47

NBER WORKING PAPER SERIES

SKILL BIASED FINANCIAL DEVELOPMENT:

EDUCATION, WAGES AND OCCUPATIONS IN THE U.S. FINANCIAL SECTOR

Thomas Philippon

Ariell Reshef

Working Paper 13437

http://www.nber.org/papers/w13437

NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue

Cambridge, MA 02138

September 2007

h b fi d f di i i h h l ll S d ill Silb


2/47

Skill Biased Financial Development: Education, Wages and Occupations in the U.S. Financial

SectorThomas Philippon and Ariell Reshef

NBER Working Paper No. 13437

September 2007

JEL No. G2,J21,J24,J3

ABSTRACT

Over the past 60 years, the U.S. financial sector has grown from 2.3% to 7.7% of GDP. While thegrowth in the share of value added has been fairly linear, it hides a dramatic change in the composition

of skills and occupations. In the early 1980s, the financial sector started paying higher wages and hiring

more skilled individuals than the rest of economy. These trends reflect a shift away from low-skill

jobs and towards market-oriented activities within the sector. Our evidence suggests that technological

and financial innovations both played a role in this transformation. We also document an increase

in relative wages, controlling for education, which partly reflects an increase in unemployment risk:

Finance jobs used to be safer than other jobs in the private sector, but this is not longer the case .

Thomas Philippon

NYU Stern School of Business

Department of Finance

44 West 4th Street, Suite 9-190

New York, NY 10012-1126

and NBER

[email protected]

Ariell Reshef

New York University

Department of Economics

19 W. 4th Street, 6FL

New York, NY 10012

[email protected]


3/47

The financial sector in the U.S. has grown steadily over the post-war period. In 1950, it

accounted for 2.3% of U.S. GDP, and 2.8% of aggregate employee compensation. In 2005,

both shares were 7.7%. These steady increases, however, hide a deep structural break, that

occurred in the early 1980s. Until then, and for more than three decades, the financial

sector had grown by hiring more employees. From the early 1980s onward, the value added

per employee increased much faster in the financial sector than in the rest of the economy,

while the number of employees remained flat or even declined.

Understanding the growth of the financial sector, and the changing composition of its

labor force, can shed light on several important economic issues. The first is the allocation

of talent in the economy. It is well understood that entrepreneurial talent is a key input

into the process leading to economic growth. Baumol (1990) argues that the allocation of

talent across occupations is more readily influenced by institutions and economic incentives

than the overall supply of talent. Economic growth requires the allocation of talent tosocially productive activities. Murphy, Shleifer, and Vishny (1991) make a similar point,

and also discuss the role of increasing returns to ability in determining the careers of talented

individuals.

Both Baumol (1990) and Murphy, Shleifer, and Vishny (1991) argue that the flow of

talented individuals into law and financial services might not be entirely desirable, because

social returns might be higher in other occupations, even though private returns are not.

Yet, there is little systematic evidence on changes in the flows of talent over time. Even if

the potential for the diversion of talent exists, it does not matter if the diversion does not

actually take place, or only on a small scale. Our paper sheds light on this issue.

Our work also contributes to the understanding of the much debated issue of skill biased

technological change and income inequality. Katz and Murphy (1992) study the secular

growth in the demand for more-educated workers from 1963 to 1987, while Autor, Katz, and

Krueger (1998), Acemoglu (1998) and Card and Lemieux (2001), among others, discuss the

role of technological improvements biased in favor of skilled workers. The finance industry


4/47

tween 1970 and 1982. Miller (1986), reflecting upon the financial innovations that occurred

from the mid 1960s to the mid 1980s, argues that the development offinancial futures was

the most significant one. Tufano (2004) argues that the more recent decades have witnessed

equally important innovations. These innovations and the development of new markets

probably changed the demand for skills in the financial sector. Moreover, they probably

increased the span over which individuals can apply their skills, making the financial sec-

tor more attractive to highly talented individuals, as emphasized by Murphy, Shleifer, and

Vishny (1991).

Finally, we contribute to the large literature on economic growth and structural change,

with the well-known contributions of Stigler (1956), Kuznets (1957) and Baumol (1967).

The rise in the finance industry is in some ways similar to the rise in other skill-intensive

services, studied recently by Buera and Kaboski (2006), but the financial sector plays a

particular role and needs to be analyzed separately. In a complete market, Arrow-Debreueconomy, there would be no financial sector. The complete market is the benchmark for

most economic applications. Studying the use of human capital in the financial sector might

inform us about the most important departures from the complete market model.

In this paper, we provide evidence on the transformation of the financial industry by

looking at the occupations, education and wages of its workforce. Was there a general

increase in the average education of the workforce entering the industry? Did some low skill

occupations disappear? Which occupations expanded? To answer these and other questions,

we turn to the Current Population Survey. Although this survey has been extensively used

to study the U.S. labor market at large, to the best of our best knowledge it has not been

utilized to study the financial sector in particular.

Our analysis uses a long time series and a representative sample of U.S. population.

This sample, however, is not appropriate for the study of very high incomes, which is the

focus of recent work by Kaplan and Rauh (2007).

We find that until the late 1970s, workers in the financial sector were only slightly more


5/47

towards investment banks, mutual, pension and private equity funds, which are relatively

more skill intensive, and an increase in skill intensity within some industries. Along with

the rise in skills, we document an increase in the wages paid to workers in the financial

sector, which is partly due to a relative increase in unemployment risk. Finally, we analyze

the changes in the set of tasks performed by workers in the financial sector.

Section 1 describes the growth of the financial sector, as well as three industries within

the industry. Section 2 considers the role of education in explaining the rise of wages paid

to employees in the financial sector. Section 3 studies the evolution of residual wages,

controlling for education, and the role of unemployment risk. Section 4 concludes. A

detailed description of the data can be found in the appendix.

1 The growth of the financial sector

In this section, we present broad trends for the U.S. financial sector over the post-war

period. We consider first the industry as a whole, using data from the Annual Industry

Accounts of the United States, and then three industries within the industry, using the

March supplement of the Current Population Survey. The data are described in details in

the appendix.

1.1 Trends for the sector

We start from the Annual Industry Accounts of the United States, published by the Bureau

of Economic Analysis. We consider three measures of the size of an industry: value added,

compensation of employees, and employment. Value added is the contribution of the indus-

trys labor and capital to the overall gross domestic product (GDP) of the country. We also

compute the compensation share. Compensation includes wage and salary accruals, as well

as supplements to wages and salaries. The compensation share and the value added share

of an industry can differ if the share of labor in the value added of the industry is not equal

to the labor share in the rest of the economy. For employment, we use the series Full-Time


6/47

Figure 1 shows that, in terms of valued added and compensation of employees, the

share of the financial sector has grown fairly linearly over the post-war period. However,

the pattern for the employment share is markedly different. The employment share of the

financial sector grows just like the value-added share until the early 1980s, but it then

flattens out and even declines somewhat after 1987.

The growth of the financial sector can therefore be decomposed into two distinct periods:

From 1947 to 1977, the share of value added of thefi

nancial sector increased from2.32% to 4.55%, an increase of 2.23 percentage points over 30 years. The share in the

compensation of employees increases from 2.72% to 4.36%. During this period, the

financial sector grows mostly by hiring more employees. The share of employment

increases from 2.39% to 3.86%, and the compensation per employee grows like in the

rest of the economy.

From the 1980s onward, the financial sector grows by increasing the value added and

compensation of its employees faster than in the rest of the economy. The employment

share remains roughly constant, increasing to a maximum of 4.64% in 1987, and then

falls to 4.32% in the early 2000s. The value added share increases from 4.55% in 1977

to 7.69% in 2005, an increase of 3.14 percentage points over 28 years, faster than in

the previous period.

1.2 Trends for industries

We now turn to the March supplement of the Current Population Survey. The Current

Population Survey (CPS) is a monthly survey of over 50,000 households conducted by the

Bureau of the Census for the Bureau of Labor Statistics. In this section, and in the rest ofthe paper, we focus on the private sector: we exclude all government employees, as well as

employees of the United States Postal Services. There are two reasons for doing so. The

first is to check that the trends identified earlier also hold for the share of finance within

h i Th d i i h h i i b


7/47

might wonder, however, whether the trends that we have described are the same in all

industries. We define three industries within the financial sector: Banking and Saving,

Insurance and Other Finance Industries. Banks, thrift and saving institutions are in-

cluded in Banking and Saving. Securities, commodities, funds, trusts, and other financial

establishments, as well as investment banks are all included in Other Finance Industries.

The top panel of Table 1 explains the classification, and Table 2 contains the number of

observations and the shares of employment of each industry on a full-time equivalent basis,

computed with the Census weighting scheme to obtain a sample representative of the U.S.

economy. Note that these are shares in the private sector, and therefore they add up to

more than the share presented in Figure 1. The trends, however, are consistent with those

from the Industry Accounts. In the CPS, the employment share of finance increases until

the early 1980s and remains constant afterwards. The wage-bill share increases linearly

throughout the 1970-2005 period, just like the value added share described above.The left panel of Figure 2 shows the size of each industry in terms of hours worked.

The share of the Insurance sector is remarkably stable, at 2% of total hours worked. The

share of Banking increases from 2% in 1970 to 3% in 1988, and then decreases to 2.75%

in 2006. By contrast, the share of Other Finance Industries was constant until 1980 at

around 0.3% before increasing to 0.9% in 2006. The right panel of Figure 2 shows the size

of each industry in terms of its wage bill. Interestingly, the wage-bill share of Other Finance

Industries increased by more than twice as much as its share of hours worked. There are

smaller differences between the (increase in) wage-bill and (flat) hour share in the other two

industries. The financial sector accounted for 4.875% of private labor income in 1967 and

8.127% in 2005, an increase of 3.252 percentage points. Other Finance Industries increased

by 1.68 percentage points, from 0.62% to 2.3%, so it accounts for slightly more than half of

the overall increase of the industry.

Result 1

Over the post war period the financial sector grew first by hiring more employees and later


8/47

An increase in the relative wage can reflect various economic forces. In the short run,

because of search costs and industry specific human capital, the supply of labor is not

perfectly elastic across industries. Thus, an increase in demand or in productivity can lead

to an increase in relative wages, holding constant the composition of the labor force. In

the long run, however, the relative labor supplies adjust and differences in wages reflect

differences in human capital and in the disutility from work in different occupations. Our

next step is therefore to investigate the education of employees in the finance industry.

2 Education of the workforce

In this section, we investigate whether the evidence in Figures 1 and 2 reflect a shift in the

composition of the labor force of the financial sector around 1980. Our CPS data cover

the period 1967-2005. Over this period, the average education of the U.S. workforce has

increased substantially. We investigate whether the improvement in education has been

relatively faster or slower in the financial sector.

2.1 Wages and education

Given that value added per employee started to increase relatively faster in the financial

sector after 1980, it is natural to study the skill composition of the labor force. To do so, we

break down hours worked between employees with less than a college degree and employees

with at least a college degree. Let hi,s,t be hours worked by employee i in sector s at time t,

and let ei,t be a dummy variable for having at least a college degree. We define the college

graduate share of hours worked in sector s at time t as:

xs,t =

Pi

hi,s,t i,t ei,t

Pi h

i,s,t i,t, (1)

where ist is the CPS sampling weight for that observation. The relative education of the

workforce of the financial sector is defined as the college graduate hour-share in Finance

and Insurance (s f in) minus the college graduate hour-share in the rest of the private


9/47

education over time. The two series display some striking similarities. Both are flat until

1980 and rise linearly afterwards, which suggests that part of the observed increase in

relative wages comes from a relative increase in the skill composition of the labor force.

Until 1980, wages are 12.5% higher and the hour-share of college graduates is 2.5 percentage

points higher in finance than in the rest of the private sector. By 2005, wages are more

than 40% higher and the hour-share of college graduates is 17.5 percentage points higher.

Result 2

The education of employees in the financial sector increased faster than in the rest of the

economy after 1980.

Does education account for the entire increase in relative wages? If wages are equalized

across sectors for a given level of education, the percentage difference in average hourly

wages should be equal to t

(xfin,t

xrop,t

) / (1 + txrop,t

), where t

is the college premium

at time t, defined as the percentage difference in the hourly wages of workers with and

without a college degree. The hour-share of college graduates increased from 13% to 30%

in the private sector between 1967 and 2005, and from 18% to 47.5% in the financial sector.

The college premium the percent difference between the wages of college graduate and

those without a college degree is around 55% in the 1970s and reaches 94% by the end of

our sample.3

Given the college premium and the differences in the hour shares of college graduates,

we would have expected finance wages to be 2.6% higher than in the rest of the economy

in 1970, while in fact they were 12.5% higher. By 2005, we would have predicted a 12.83%

difference in wages, compared to the observed 40%. The discrepancy comes from the fact

that employees of the financial sector earn more than employees in other sectors with similarlevels of education, and that the gap has increased over time. We address this issue in section

3.

2.2 Skill intensity


10/47

s is:

ys = f1/s h11/ + (1 s)1/ l11/; Xs , (2)where Xs is a vector of industry characteristics, and h and l are hours worked by high

education and low education workers, respectively. This production function has a constant

elasticity of substitution between high and low skill labor, . Many models of macroeco-

nomic growth use production functions with constant values for . Let wh and wl be the

hourly wages for high and low education workers. Cost minimization of the wage bill for

any level of output in (2) implies that the relative demand for high skill labor must satisfy:4

hsls

=s

1 s

whwl

. (3)

In the benchmark case, where and wl/wh are the same across sectors, the relative skill

intensity hs/ls depends only on the relative parameters s across different industries.5 In

particular, using (3) to compare finance versus the rest of the economy leads to:

hfin/lfinhrop/lrop

=fin

1 fin

1 roprop

. (4)

Figure 4 focuses on employees with a least a college degree. The left panel shows the

share of hours worked by employees with at least a college degree in each industry, computed

according to equation (1). For the total private sector, it increased from 13.2% in 1968 to

30.6% in 2006, with a sample average of 23%. This share, however, varies a lot across sectors.

The sample average is 53.8% for Other Finance Industries, 32% in Insurance companies,

and 26.3% in Banking. The trend increase in Insurance is the same as in the rest of the

private sector, but the trends are steeper in Banking and Other Finance Industries. The

share of college graduates in Banking was roughly the same as in the rest of the private

sector in the early 1970s (even slightly lower), but 10 percentage points higher in 2005.

The right panel of Figure 4 shows the evolution of the left-hand side of equation (4)

over time, using hours worked by college graduates as a measure of h. The series are

normalized to have a mean of one over the first five years 1967 to 1971 Interestingly the


11/47

relative skill intensity of Insurance has remained roughly constant throughout the sample

period, consistent with the benchmark model and balanced growth. On the other hand, the

relative skill intensity in Banking and in Other Finance Industries has increased sharply.

The Banking sector was slightly less skill intensive than the rest of the economy in the early

1970s, but it became significantly more skill intensive over the 1980s. The Other Finance

Industries sector was already skill intensive and has become more so over time.

Result 3The relative skill intensity of the Insurance industry has remained approximately constant.

In contrast, Banking and Other Finance Industries have become more skill-biased than the

rest of the economy after 1980.

So far, we have documented an increase in the skill intensity of the financial sector

relative to the rest of the economy, and a change in the allocation of the workforce withinthe financial sector away from Banking and towards Other Finance Industries. These trends

presumably reflect changes in the set of tasks performed by workers in the financial sector,

a topic that we now discuss.

2.3 Occupations

Our classification of occupations attempts to group employees according to the tasks they

perform, irrespective of the industry in which they work. For instance, we classify all em-

ployees trading securities as traders, no matter whether they work in Banking, Insurance

or Other Finance Industries. We do the same for managers and for administrative work-

ers. Unfortunately, it is hard to find consistent definitions of occupations over time. The

appendix explains in details what we did, the constraints we faced and the reasons for our

choices. At the end of the day, we use six occupations, presented in the bottom panel of

Table 1. Table 3 contains the number of observations for each occupation and each year,

and the shares of each occupation in the total employment of the Finance and Insurance


12/47

This is true in all industries, and it is particularly important in Banking. The decrease in

the share of clerks explains much of the evolution of average wages in Banking after 1980.

Securities and Financial Asset Sales and Computer and Mathematics are the fastest

growing occupations.6 It is clear that the financial sector has become more trading-intensive

and more computer-intensive over time.

Result 4

Much of the skill bias in Banking reflects the decline of bank tellers and other administrative

occupations. The fastest growing occupations infinance are related to the trading offinancial

assets, and the use of computers and mathematics.

2.4 Technological or financial innovations?

Autor, Levy, and Murnane (2002) sort jobs and tasks along two dimensions: manual orcognitive, routine or non-routine. Computers can perform routine tasks and are therefore

substitutes for labor employed in routine jobs. Computers are complements to non-routine

tasks, especially cognitive ones, and they are neither complements nor substitutes to manual

non-routine tasks. As a result, employees in abstract or analytical tasks become more

productive, the demand for routine jobs decreases, while manual jobs are less affected.

The evolutions of the tasks performed by the financial sector provide support for the

idea that advances in information technology, and in particular the decrease in the price of

computing power, have considerably decreased the demand for routine jobs and increased

the productivity of non-routine abstract jobs.

The relative stability of Insurance, however, suggests that financial innovations might

also have played a role. Insurance was already a skill intensive industry in 1980, yet its

evolution does not suggest a strong skill bias afterwards. Moreover, one might think that

improvements in computers by themselves affected the insurance sector as much as the other

financial sectors.


13/47

Silber (1983), only 2 or 3 are related to Insurance. These innovations had a larger impact

on Other Finance Industries than on the Insurance industry. Many of these innovations are

market-oriented, which explains the nature of the fastest growing occupations in Figure 5.

This is also consistent with the argument in Miller (1986) on the importance of financial

futures markets.

3 Wage premia in the financial sector

To what extent can education account for the change in relative wages that we have doc-

umented? In this section, we document that even controlling for education, finance wages

have increased faster than in the rest of the economy, and then we show that this might

partly reflect an increase in unemployment risk. To make the discussion more concrete, we

start by considering the striking example of engineers.

3.1 Financiers and engineers

In this section, we focus on a particular comparison, that between engineers and financiers

with similar levels of education. Figure 6 shows the evolution of the average annual wages

of employees in the Finance and Insurance sector with a college degree or more, and of

engineers employed in the private sector (but not in Finance and Insurance). The wages

are in constant 2000 prices. The left panel uses individuals with exactly a college degree,

and the right panel uses employees with a post-graduate degree. In both cases, the relative

wage offinance employees was constant until the 1980s, and then started to increase faster

than the wage of engineers with the same level of education. The picture is particularly

striking for post-graduates, a category that includes MBAs and PhD graduates.

Moreover, the CPS underestimates the income of individuals who earn very high salaries,

due to top-coding of income. Therefore, the wages that we report may not be accurate for

certain occupations, Securities and Financial Asset Sales in particular. In our sample, the

percent of top-coded observations in the private sector is around 1%, but it is around 2%


14/47

3.2 Wage regressions

We have seen that the wages of employees in the financial sector have increased relativeto the rest of the private sector. This increase has been accompanied by an increase in

the share of college graduates working in the financial sector. The first question we seek

to answer is whether the change in the composition of the labor force, in particular its

education, can account for all of the observed increase in wages.

To answer this first question, we run a series of cross-sectional regressions on sub-

samples offive years.7 Pooling the data ensures that we have enough observations in each

cell. We restrict our attention to full time employees of the private sector, i.e., employees

who work at least 40 weeks a year, and at least 35 hours a week. 8 For each subsample

{[1967, 1970] ,... [2001, 2005]}, we run the following regression:

log(wit) = t, + Xi,t + edui,t + 1

it

+ uit, (5)

where t . In equation (5), wit are the annual wage earnings of individual i in year

t, Xit are individual characteristics of individual i at time t (race, sex, marital status,

urban residence, potential experience and its square), and eduit is a vector of dummies for

education groups: 12 years of education, 13 to 15 years, 16 to 17 years, and 18 years or

more. The variable 1

it is a dummy, equal to one if individual i works in thefi

nancial sectorin year t. Within each period we also control for year effects t,. We allow the vector of

coefficients to vary over periods in order to capture changes in the returns to education

common to all sectors in the economy. The coefficient measures the extent to which

every finance employee receives a higher wage, controlling for all other things. It captures

compensating wage differentials or unobserved heterogeneity.

Table 4 displays the results of the regressions (5). The most striking feature of the data

is the increase in the returns to education, especially for high education. Mincer (1997),

Lemieux (2006) and Deschnes (2006) show that log wages have become an increasingly

convex function of the years of education since the 1970s, and Autor, Katz, and Kearney


15/47

has increased dramatically, while the gap between high-school graduates and dropouts has

increased only moderately. Controlling for education is important if one wants to understand

the evolution of the wages of workers in the financial sector.

Figure 7 plots the evolution of from two specifications of equation (5), one where we

exclude education (not reported here) and one where we include education, which corre-

sponds to the first row of Table 4. The drop from one line to the other shows that education

accounts for six to eight percentage points of the difference in log wages. However, even

controlling for education, we find that individuals working in the Finance and Insurance

sector earn more than individuals working in the Rest of the Private Sector. The finance

premium was small, around 3.4% until the mid-seventies. It started to grow in the late

1970s, to reach 20% in the early 2000s. To better understand the finance premium, we

decompose it into three premia, one for each of our industries. Figure 8 shows that the wage

differentials have increased in all three industries. In Banking, there is no wage differential

in the early 1980s. The wage differential is much larger in Other Finance Industries, which

is the more skill intensive sector. Moreover, in this industry it increases the most, by more

than 20 percentage points. However, since we control for observed skill, this is indeed a

premium for working in the sector.

Our previous specification implicitly assumed that wage differentials were the same for

all education groups. We now relax this assumption and run the following regression:

log(wit) = t, + Xi,t + edui,t +

+ edui,t

1it + uit.

The vector of coefficients measures the additional return to education in the financial

sector. Table 5 displays the results of the regressions and Figure 9 plots the evolution of the

wage earnings of college and high school graduates inside and outside finance. Within each

education group, the wage differential is small in the 1970s and starts to increase in the early

1980s. The wage regressions therefore confirm that a specific transformation took place in

the financial sector in the late 1970s and early 1980s one that affected all education groups

and all industries within finance In addition the results show that the college premium


16/47

Controlling for the observed characteristics of employees, the relative wage infinance started

to increase in the late 1970s. A similar pattern emerges within each industry and within

each education group.

3.3 Wage differentials and unemployment risk

That two individual with the same education and observable characteristics earn such differ-

ent income can be explained in one of three ways: compensating differentials, employment

and wage risk, and unobserved heterogeneity. According to the first candidate explanation,finance jobs are relatively less attractive than non-finance jobs, and, in equilibrium, this

is reflected in the relative wages. Compensating differentials are certainly important, and

they can account for a fraction of the average differences that we observe. However, they

do not strike us as a plausible explanation for the increase in the relative wages paid in the

financial sector, because it is hardly sensible to argue that finance jobs are less interesting

today than they were 30 years ago, especially relative to the rest of the private sector.

Another, and in our view more plausible, explanation for the trends in relative compen-

sation, is a relative increase in labor income risk in the financial sector. Iffinance workers

are more likely to loose their jobs, or if their wages are more volatile, they would have to

be compensated for this extra risk. To test this explanation, we fit two regressions. Let it

be a dummy for being employed at time t. First, we run a logit regression on the likelihood

of becoming unemployed:

Pr (it+1 = 0 | it = 1) = f

1it; wit; edui,t; Xi,t

(6)

where f is the logistic function.

The estimation of equation (6) requires a longitudinal dimension. Therefore we use theMatched CPS, which allows us to potentially observe each individual in the CPS twice,

in two consecutive years. See the appendix for a complete description of the methodology

involved in matching observations on individuals from consecutive surveys.


17/47

to be in the sample at time t, an individual must be in the labor force, and employed in

the private sector. At time t + 1, the individual must be in the labor force (in the private

or public sectors), either employed or not. On average, over the whole sample, 6.91% of

individuals experience a transition from employment to unemployment. Individuals in the

financial sector are less likely to loose their jobs: only 4.39% make the transition. The

relative stability offinance jobs has decreased over time, however, as one can see in Table 6,

which reports our estimates for equation (6). As expected, full time, married, experienced

workers, and workers with a higher level of education, are less likely to become unemployed.

Figure 10 shows the evolution of unemployment risk in the financial sector relative to

the private sector. The probability of becoming unemployed is evaluated for the average

worker, i.e., it is evaluated at the means of all other variables. From the late 1970s to the

early 1990s, the stability of jobs in the financial sector has converged to the average for the

rest of the private sector. This convergence started in the early 1980s, which coincides with

the timing of the increase in relative wages in finance, as shown in Figure 2.

When we estimate equation (6) with separate dummies for the three industries, we find

clear trends in Banking and Insurance. In Other Finance Industries we do not observe

enough transitions to draw reliable conclusions. We have also estimated an equation with

wage risk, measured either by the volatility of the log wage, or by the probability of a drop

of more than 10%, on the left hand side. We did not find any evidence that wage risk is

higher in the financial sector, or that the relative risk has changed over time.

An important concern with our controls for education is that it might not capture

important heterogeneity among individuals with the same number of years of schooling.

A better way to capture this heterogeneity is to control for the current wage, rather than

education. To do so, we construct three wage groups with cutoffs at the 34th and 67thpercentiles of the overall wage distribution in a given year. Note that the cutoffs correspond

to the distribution of wages in the entire private sector, not the financial sector, because the

correct experiment is to compare individuals with the same wages inside and outside the


18/47

Result 6

Unemployment risk was lower in the financial sector than in the rest of the private sector

until 1980. It increased in relative terms over the 1980s and early 1990s.

A simple calibration

Based on the evidence presented so far, we can propose a first interpretation of the data.

Regarding the level of compensation, a constant compensating differential appears to be

required, since even in the more recent years, the unemployment risk in the finance industry

is not higher than in the rest of the economy. It has merely converged to the same level.

The increase in the relative unemployment risk in the financial sector can however account

for some of the increase in relative wages. Ruhm (1991) finds that layoffs lead to temporary

unemployment and long lasting decreases in earnings: Displaced workers were out of work

eight weeks more than their observably similar counterparts in the year of the separation,

four additional weeks in period t+1, and two extra weeks at t+2. By year t+3 they were

jobless only 1.5 weeks more than the peer group, and the t+4 increase was just six days.

By contrast, almost none of the t+1 wage reduction dissipated with time. The earnings

gap remained at 13.8 percent and 13.7 percent, respectively, in years t +3 and t +4.

A complete study of the effects of unemployment risk on the level of compensation that is

needed to keep workers indiff

erent between diff

erent jobs is clearly beyond the scope of thispaper. Nonetheless, we think it is useful to provide some simple benchmark calculations.

We do so in the simplest framework possible and we assume that labor income is the only

source of risk and that the utility function has constant relative risk aversion. We set the

personal discount rate and the market rate both equal to 3% per year. We assume that

workers live and work for 40 years, and that the labor income process, yt, is given by

yt+1 =

1.02 yt with probability 1p

0.86 yt with probability p

, and y1 given.

The increase of 2% captures the normal increase in real labor income. The drop by 14%

captures the income loss from displacement documented by Ruhm (1991). This process


19/47

increase the unemployment risk to p = 6.91%. This increase of 2.5 points corresponds to

the increase in relative unemployment risk that we have documented earlier. In order to

keep workers indifferent, the new starting wage should be y1 = 1.084, an increase of 8.4%.

If we lower the calibrated risk aversion to 1, the required increase in wages is 7.8%. If we

increase risk aversion to 3, the required increase in wages is 9%. Recall that the premium

paid to finance employees has increased by 16.6 points, from 3.4% to 20%. Thus, it appears

that the increase in unemployment risk could account for half of the increase in relative

wages.

4 Conclusion

Until the late 1970s, workers in the financial sector were only slightly more educated and

received slightly higher wages than in the rest of the economy. Since the early 1980s,

however, the financial sector has been hiring more and more skilled individuals. There hasbeen a composition shift away from Banking and towards relatively more skill intensive

industries. In addition, there has been an increase in skill intensity within Banking and

within Other Finance Industries.

Our results suggests two sources of skill bias: computers and market-related financial

innovations. First, there is clear evidence that jobs involving routine tasks have tended

to disappear, probably because of improvement in computers and information technologies.

This is especially true in Banking, where the trend may have been reinforced by deregulation

and consolidation. The relative stability of Insurance, however, as well as the fact that the

fastest growing tasks involve the trading of financial assets suggests that innovations in

financial markets also played an important role.

Controlling for education and observable characteristics, we find that employees of the

financial sector earned 3 to 4% more than employees in the rest of the private sector during

the 1970s. At that time, however, they also enjoyed substantially lower unemployment

risk. After 1980, the unemployment risk in the financial sector started to catch up with


20/47

for one half of the increase in the finance wage premium. The remaining half probably

reflects unobserved heterogeneity in the labor force.

Overall, the share of talented individuals hired by the financial sector has increased

very significantly over the past three decades. This finding raises several questions for

future research. First, does the increase in skilled labor in the financial sector lead to

more innovations in this sector? Second, how do financial innovations affect the rest of

the economy? Finally, what are the welfare consequences of the shift in the allocation of

talent?1010 Philippon (2007) discusses the optimal allocation of talent in a model with credit constraints and indus-

trial innovations, but no financial innovations.


21/47

Appendix

A Description of the data

A.1 Annual industrial accounts

We obtain data on employment and value added at the industry level from the Bureauof Economic Analysis (BEA). Value added, the contribution of each industrys labor andcapital to the overall gross domestic product (GDP), is equal to an industrys gross output(Sales or receipts and other operating income, commodity taxes, and inventory change)minus its intermediate inputs (consumption of goods and services purchased from other

industries or imported). Current-dollar value added by industry is calculated as the sumof disbursements to its labor and capital which are derived from the components of grossdomestic income.

Employment could be defined in a number of ways: all employees, full-time equivalent,or total hours worked. There is, however a compatibility issue around 1997, when the BEAchanged its industry classification from SIC to NAICS. As a result, only the series for thenumber of full time and part time employees (FTPT) is available consistently, and this isthe one that we use. The 1948-97 full-time and part-time employees was released in Octo-ber 2006, by the Industry Economic Accounts, Bureau of Economic Analysis (BEA), U.S.

Department of Commerce. Because this series is based on the current NAICS classification,it is possible to link the 1948-1997 and 1998-2005 samples.

A.2 The Current Population Survey

Our data on individuals comes from the March supplement of the Current Population Sur-vey (Annual Social and Economic Study) from survey years 1968-2006, which pertain to1967-2005 actual years. A CPS year refers to data of the preceding year, i.e. March CPS2006 documents annual data from calendar year 2005. We therefore adopt the following

taxonomy: We call year the actual year that the survey pertains to, while a CPS year isdenoted as survey year. The Current Population Survey (CPS) is a monthly survey ofabout 50,000 households conducted by the Bureau of the Census for the Bureau of LaborStatistics.11 Currently, there are more than 65,000 participating households. The sample isselected to represent the civilian non-institutional U.S. population. The CPS includes dataon employment, unemployment, earnings, hours of work, and other demographic character-istics including age, sex, race, marital status, and educational attainment. Also availableare data on occupation, industry, and class of worker. We choose to use only one particularmonth survey, the March supplement, for two reasons. First, this supplement contains more

demographic details, in particular on work experience and income sources and amounts.Since 1976, the survey has also been supplemented with a sample of Hispanic households(about 2,500 interviewed). Second, it has been extensively used in the empirical labor andmacro-labor literature, which lends to the comparability of our results. Let us now definethe groups that we use in our empirical analysis. We restrict attention to individuals whoare in the labor force, of at least 15 years of age.

I d t Cl ifi ti


22/47

trusts, and other financial investments as well as investment banks are all included in OtherFinance Industries. Table 2 shows the number of observations of employees surveyed fromthe three financial industries and from the rest of the private sector from 1968 to 2006.Table 2 also shows the shares of employment, computed with the Census weighting schemeto obtain a sample representative of the U.S. economy. These sectors are consistentlyidentified, without any "jumps" or "drops" in their shares of total employment, despitechanges in industrial classifications in the CPS in our sample, which occur following eachdecennial census. The major industrial re-classifications occurred in survey year 1983, fromthe Census 1970 system to the 1980 system; and in survey year 2003, from the Census 1990system to the 2000 system. Of these two re-classifications, the latter was more substantialoverall, yet it does not affect our sectors. The Census Bureau provides industrial crosswalksfor the 1970-1980 systems and for the 1990-2000 systems, from which one can gauge howsome industries are split or merged into others (Census Bureau (1989), Census Bureau(2003)). These crosswalks are basically a transition matrix for all industries from oneclassification to the other. A close examination of these transition "probabilities" lead usto conclude that our industries are consistently defined throughout our sample. In thetransition from the 1970 system to the 1980 system 99.9% remain inside each industry; andfor the transition from the 1990 system to the 2000 system over 95% of workers remaininside each industry. This is due to the fact that the functions of our three industries arenarrowly and well defined, and due to the fact that they are not too large.

Occupations

Examining the distribution of occupations within our three industries lead us to choosesix occupation groups (henceforth, "occupations"), which describe the major occupationalgroups in our sample. These are: Managers and Professionals, Computer and Mathe-matics, Insurance Specialists, Securities and Financial Asset Sales, Administration,Including Clerks, and All the Rest (janitors, security and miscellaneous). As with indus-try classifications, major occupational re-classifications occurred in survey year 1983, fromthe Census 1970 system to the 1980 system, and in survey year 2003, from the Census 1990system to the 2000 system. Of these two re-classifications, the latter was more substantial.We examined the occupational crosswalks, which are provided by the Census Bureau tomake sure that our occupational groups are consistently defined over time (Census Bureau(1989), Census Bureau (2003)). We could not consistently separate "managers" from "pro-fessionals" due to re-classifications in survey years 1983 and 2003; some occupations thatwere defined as "professional" were split and re-classified as "managerial" and vice versa.However, these two groups together are consistently identified, without any "jumps" or"drops" in their employment shares over time. Much effort was devoted to making surethat the other occupation groups are also consistently defined throughout our sample. Seethe appropriate appendix for a detailed description of occupation groups and how theyare constructed. Note that some of these occupations potentially mean different things indifferent industries. For instance, in Banking the Managers and Professionals includebank officers, but these officers do not exist in the two other industries. Another examplefrom Banking is that bank tellers are a significant part of the Administration, IncludingClerks, but, once again, they do not exist in other industries. However, our more narrowlydefined occupations, Computer and Mathematics, Insurance Specialists and Securitiesand Financial Asset Sales are the same in all industries. Table 3 contains the number of


23/47

year 1991 we correct years of schooling for individuals who did not complete the last yearin school by subtracting one year. This correction is not needed after survey year 1992.From survey year 1992 and on early school attainment is lumped into groups: 0 years, 1-4years, 5-6 year and 7-8 years. Also starting in survey year 1992 school attainment startingwith high school is marked by degrees, not years, therefore it is not possible to distinguishbetween, e.g., 13, 14 and 15 years of school. To make our education variable consistentthroughout our sample, we adopt the coding that starts in survey year 1992, i.e., we groupearly school attainment into brackets for all the sample and assign maximal values to eachbracket. Also, we group 13, 14 and 15 years of school together and assign 14 years for allindividuals within that bracket in all years. In addition, we lump 17 years of schoolingtogether with 16 years, for similar reasons. This makes the education variable categoriessmooth throughout the sample, and in particular around the 1991-1992 surveys. In ouranalysis we consider mainly two education categories: less than 4-year college graduates(up to 15 years of school, including 2-year college graduates), and college graduates (16years of school, including 4-year college graduates, and post-graduates). These groups aresufficient to characterize differences in skill. In some cases where it is not, we separatetertiary education into only college and post-graduate degrees. Experience is potentiallabor market experience. It is measured as min(age-edu-6;age-18), where edu is years ofschooling. The CPS does not contain data on job spells.

Annual hours workedIn order to compute annual hours worked we multiply the number of weeks worked byaverage hours worked in a week. Weeks worked are grouped into 7 brackets before surveyyear 1976. Therefore we impute the average number of weeks worked before survey year1976 by calculating the average weeks worked within those brackets in later years, by sex(male\female). By this we get a better estimate of typical number of weeks worked for eachsex in earlier years. Average hours worked per week are reported from survey year 1976 andon, but not beforehand. Until survey year 1975 only hours last week are reported, so we usethose when available. For individuals who worked during the survey but where hours lastweek are not reported we impute average hours worked by calculating the average hoursworked per week in later years by sex and full-time\part-time status (full-time means atleast 35 hours a week).

Wages and top-codingWe deflate all wages reported in the CPS using the deflator for personal consumptionexpenditures from the Bureau of Economic Analysis. The reference year is 2000. Hourlywages are calculated by dividing annual wage income by number of hours worked. The CPSunderestimates the income of individuals who earn very high salaries, due to top-coding ofincome. Therefore, the wages that we report may not be accurate for certain occupations,Securities and Financial Asset Sales in particular. In our sample, the percent of top-codedobservations in the private sector increases from 0.06% in 1967 to 1.1% in 1980, after whichit fluctuates in the range 0.38%-1.6%, due to secular adjustments of the top-coding incomelimit. However, in the financial sector there are many more incidents of top-coding: inBanking, there are on average twice as many top-coded observations, in Insurance there areon average 2.4 as many top-coded observations, whereas in Other Finance Industries thereare on average 13 times as many top-coded observations. This leads to an under-estimationof relative wages in the financial sector. In an attempt to compensate for this, we multiply


24/47

B Construction of occupation cells

In order to create meaningful, but not-too-small occupation cells, we first tabulated all oc-

cupations in each of the three industries, Banking and Saving, Other Finance Industriesand Insurance separately over time. Merging the industrial classifications for these threeindustries over time was relatively straightforward, since changes in industrial classificationsover the years did not result in any significant re-allocation of workers in these sectors. Seedescription in the text above. The 2-digit occupation classifications did not contain enoughinformation on some occupations of interest, so the tabulation was done at the 3-digit level(the most detailed available). Consistent classifications are available within the followingsurvey year intervals: 1968-71, 1972-82, 1983-2002, 2003-06. For 1968-71, the census 1960classification was used, for 1972-82 the Census 1970 classification, for 1983-2002 the Census

1980 and 1990 classifications, and for 2003-06 the Census 2000 classification. Census 1980and 1990 classifications are virtually identical, but there are significant re-allocations ofoccupations between the 1970 and 1980 systems, and between the 1990 and 2000 systems.Since the classifications change from time to time, much effort went into identifying therelevant classifications across years. This was done by comparing high-frequency occupa-tions from the tabulation with their headings from the CPS documentation. Then, usingthe apparent occupation groups, time series of occupational shares within each industrywere graphed in order to make sure that the groups are consistent across time. Major re-classifications occurred in survey year 1983 and in survey year 2003; in these cases we checked

the occupational crosswalks from the Census Bureau to make sure that our occupationalgroups are consistently identified (Census Bureau (1989), Census Bureau (2003)). Finally,we defined six occupation groups (henceforth, "occupations"), which concisely describe themajor occupational groups in our sample in a parsimonious way. These are: Managersand Professionals, Computer and Mathematics, Insurance Specialists, Securities andFinancial Asset Sales, Administration, Including Clerks, and All the Rest (janitors,security and miscellaneous). Some additional adjustments were done by comparing averagewages and education levels within 3-digit occupation cells to average wages in our occupa-tional groups. This helped us understand better the nature of some occupational headings

and to hint towards where they belong within our occupational groups. In particular, thisled us to allocate "supervisor-of..." occupations in survey years 1983-2002 into "Managersand Professionals", since their wages were similar to that group and an order of magnitudehigher than closely-labeled occupations. But we did not allocate "first-line supervisors" inCPS 2003-06 into that group, as their wages did not conform to the previous rule.

Managers and ProfessionalsManagers are quite heterogeneous. They include chief officers, but also middle manage-ment, operation managers, financial managers, human resources managers, etc.. We founda large range of variation within this group in terms of education. Management-related

occupations are part of the managerial occupation classifications in survey years 1983-2002,but are mostly allocated to other occupations in the Census 2000 system and 1970 system.Therefore we allocated managers with professionals in all years. A close examination ofthe titles and wages of these occupations justifies this allocation. Professionals are alsoquite heterogeneous in terms of occupation titles, but less so in terms of education lev-els. The most prevalent occupations in this cell are economists, accountants, analysts and


25/47

officers cannot be separately identified before 1983so we aggregated managers and pro-fessionals together as one unit throughout the sample in order to keep a consistent group.When keeping these two groups separate, the real wages of each group separately seem to

be consistent over time and do not exhibit any jumps due to the change in classificationsystems aforementioned. In particular, wages of professionals do not jump upward after1983, when bank officers are allocated to that group. Managers wages are somewhat higherthroughout, and start increasing more after 1985. But the wages of Managers and Profes-sionals in Banking commove quite closely; this might reflect the relatively tight hierarchicalorganizational structure in Banking.

Computer and MathematicsThis cell is quite homogeneous throughout. Much effort was put into making it consistentthrough time. Despite significant variance in distribution across education categories, indi-

viduals in this cell earn relatively similar wages. This indicates that tasks\productivity inthis group do not vary much across education categories. The cell includes computer sci-entists and systems analysts, computer programmers, software engineers, database admin-istrators, network and computer systems administrators, etc.. In addition, it also includesmathematicians. The cell does not include actuaries and statisticians for the simple reasonthat these two occupations appear almost only in Insurance and, therefore, are allocated tothe professionals cell.

Insurance SpecialistsThis cell includes only Insurance agents, adjusters, examiners, investigators and underwrit-

ers. Almost all observations in this cell come from the Insurance industry, but there is asignificant number of observations in Banking and Saving. Insurance adjusters, examiners,and investigators were split from 2003 CPS into two categories (Claims adjusters, apprais-ers, examiners, and investigators and Insurance claims and policy processing clerks). Were-allocated them together to get a consistent group.

Securities and Financial Asset SalesThis cell includes stock, bond, commodity and other asset traders, and personal financialadvisors. Almost all observations in this cell come from the Other Finance Industries in-dustry, but there is a significantand growing number of observations in the other two

industries. Before survey year 1971 (i.e., 1968-70) many individuals that would fit in Secu-rities and Financial Asset Sales were actually allocated to broad managerial classificationsand cannot be separated.

Administration, Including ClerksThis cell is probably the most diverse in terms of education levels and wages. It containsa large range of different occupations that operate in offices, like tellers, clerks, secretaries,office assistantsbut does not include managers thereof. In the Banking industry this cellincludes a large number of bank tellers. In all industries the cell includes a large number ofsecretaries.

C Construction of Matched CPS

We thank Donghoon Lee for providing us with his methodology. The "Matched CPS" takesadvantage of the fact that households in the CPS are sampled for more than a year, in thef ll i E h h h ld h h i h i l d f


26/47

Unfortunately, the CPS does not hold a definitive person ID, by which one could easilymatch two observations on the same individual from two consecutive surveys. The followingmethodology is used to match observations on the same individual from two consecutive

surveys. We match individual observations from two consecutive surveys by household ID,their "line" within the household (which is an intra-household identifier), state of residence,race, sex and year of birth. These are supplemented with a few more identifiers generatedby the CPS (segment number, serial number and a random cluster code). We make surethat there are only two observations within each cell defined by these identifiers and dropall other cells.

Some survey years cannot be matched. Survey year 1968 cannot be matched backwards,because our sample starts with that survey year. Likewise, survey year 2006 cannot bematched backwards, because our sample end with that survey year. Other survey years

that cannot be matched for technical reasons are 1971, 1972, 1976, 1985, 1995 and 2001.Approximately 63% of all observations are actually matched from within survey years thatcan be matched.

Definition of unemploymentHere we give the exact definition of our unemployment indicator. A person would get apositive indication of unemployment if:

1. did not work last year and reported: could not find work, looking for work or on layoff.

2. in survey years 1968-1993 major activity in the week before the survey was lookingfor work.

3. in survey years 1968-1993 did not work last week due to being laid-off.

4. in survey years 1994-2006 reported being on layoff or looking for work.

5. in survey years 1968-1988 reported reason for working part year was looking for workor being unemployed.

6. reported positive number of weeks looking for work last year.

7. reported positive number of weeks being unemployed last year.

Since the sample for our transition regressions includes only people that were not unem-ployed in the first year they were surveyed, this eventually reduces our sample. Table Acontains the number of observations in the transition regressions for every year that couldbe matched.


27/47

References

Acemoglu, D. (1998): Why Do New Technologies Complement Skills? Directed Tech-

nical Change and Wage Inequality, The Quarterly Journal of Economics, 113(4), 10551089.

Autor, D., L. Katz, and A. Krueger (1998): Computing Inequality: Have ComputersChanged the Labor Market?, Quarterly Journal of Economics, 113(4), 11691214.

Autor, D. H., L. F. Katz, and M. S. Kearney (2006): The Polarization of the U.S.Labor Market, American Economic Review, 96(2), 189194.

Autor, D. H., F. Levy, and R. J. Murnane (2002): Upstairs, Downstairs: Computersand Skills on Two Floors of a Large Bank, Industrial and Labor Relations Review, 55,432447.

Baumol, W. J. (1967): Macroeconomics of Unbalanced Growth: The Anatomy of theUrban Crisis, American Economic Review, 57, 415426.

Baumol, W. J. (1990): Entrepreneurship: Productive, Unproductive, and Destructive,Journal of Political Economy, 98, 893921.

Buera, F. J., and J. P. Kaboski (2006): The Rise of the Service Economy, WorkingPaper, Northwestern University.

Card, D., and T. Lemieux (2001): Can Falling Supply Explain the Rising Return toCollege for Younger Men? A Cohort-Based Analysis., Quarterly Journal of Economics,116(2), 705746.

Census Bureau (1989): The Relationship Between The 1970 and 1980 Industry andOccupation Classification Systems, Discussion Paper Technical Paper 59.

(2003): The Relationship Between the 1990 Census and Census 2000 Industryand Occupation Classification Systems, Discussion Paper Technical Paper 65.

Deschnes, O. (2006): Unobserved Ability, Comparative Advantage and the Rising Re-turn to Education in the United States 1979-2002, Working Paper, University of Cali-fornia, Santa Barbara.

Kaplan, S. N., and J. Rauh (2007): Wall Street and Main Street: What Contributesto the Rise in the Highest Incomes?, NBER Working Paper 13270.

Katz, L. F., and K. M. Murphy (1992): Changes in Relative Wages, 1963-1987: Supplyand Demand Factors, Quarterly Journal of Economics, 107(1), 3578.

Kuznets, S. (1957): Quantitative Aspects of the Economic Growth of Nations: IndustrialDistribution of National Product and Labor Force, Economic Development and CulturalChange, 5, 1111.


28/47

Mincer, J. (1997): Changes in Wage Inequality, 1970-1990, in Research in Labor Eco-nomics, ed. by S. W. Polachek, pp. 118, London. JAI Press.

Murphy, K. M., A. Shleifer, and R. W. Vishny (1991): The Allocation of Talent:Implications for Growth, The Quarterly Journal of Economics, 106, 503530.

Philippon, T. (2007): Financiers versus Engineers: Should the Financial Sector Be Taxedor Subsidized?, Working Paper, NYU.

Ruhm, C. J. (1991): Are Workers Permanently Scarred by Job Displacements?, TheAmerican Economic Review, 81(1), 319324.

Silber, W. L. (1983): The Process of Financial Innovation, The American Economic

Review, 73(2), 8995.

Stigler, G. J. (1956): Trends in Employment in the Service Industries, General Series.NBER.

Tufano, P. (2004): Financial Innovation, in The Handbook of the Economics of Finance,ed. by M. H. George Constantinides, and R. Stulz. North Holland.


29/47

Year Matched observations Year Matched observations

1967 14,318 1986 17,441

1968 13,683 1987 16,5681969 15,004 1988 17,555

1970 . 1989 19,146

1971 . 1990 18,393

1972 9,299 1991 17,634

1973 13,882 1992 13,497

1974 8,857 1993 12,704

1975 . 1994 .

1976 15,761 1995 17,461

1977 15,984 1996 17,794

1978 16,516 1997 17,639

1979 20,268 1998 17,9791980 18,014 1999 18,406

1981 17,519 2000 .

1982 16,525 2001 25,253

1983 16,256 2002 25,390

1984 . 2003 22,162

1985 16,899 2004 23,498

TOTAL 547,305

Appendix Table A: Matched CPS Sample

Notes: This sample takes into account the number of successful matches in the

CPS and the following restrictions. Individuals in the first matched year must be in

th l b f l d i th i t t i th d t h d


30/47


31/47

Non Financial

Private Sector

Year Obs. Obs. Empl. Share Obs. Empl. Share Obs. Empl. Share

1967 56002 1046 0.0181 149 0.0026 1008 0.0173

1968 57229 1039 0.0176 197 0.0033 1001 0.0168

1969 55806 1118 0.0193 227 0.0040 1030 0.0179

1970 56644 1098 0.0188 196 0.0033 1140 0.0195

1971 54208 1049 0.0186 167 0.0030 1121 0.0198

1972 53218 1112 0.0201 190 0.0034 1048 0.0189

1973 53695 1105 0.0198 183 0.0033 1045 0.0187

1974 52654 1217 0.0220 142 0.0026 1056 0.0191

1975 53979 1154 0.0205 154 0.0028 1009 0.0181

1976 65773 1382 0.0203 179 0.0029 1124 0.0171

1977 64855 1394 0.0211 171 0.0028 1128 0.0178

1978 65574 1514 0.0221 185 0.0030 1216 0.0190

1979 77736 1812 0.0228 207 0.0030 1408 0.0185

1980 77096 1919 0.0249 237 0.0035 1452 0.0195

1981 68789 1749 0.0245 267 0.0041 1294 0.0189

1982 67709 1708 0.0244 276 0.0045 1250 0.0186

1983 67173 1732 0.0250 290 0.0045 1293 0.0196

1984 68834 1840 0.0256 361 0.0055 1253 0.0180

1985 67826 1780 0.0254 318 0.0047 1343 0.0193

1986 67575 1798 0.0254 357 0.0054 1275 0.0183

1987 67589 1955 0.0275 408 0.0057 1400 0.0201

1988 63617 1752 0.0266 350 0.0057 1314 0.0201

1989 69460 1923 0.0260 434 0.0062 1474 0.0205

1990 68970 1842 0.0249 344 0.0049 1491 0.0208

1991 68015 1820 0.0246 331 0.0050 1400 0.0204

1992 67035 1746 0.0247 377 0.0059 1286 0.0187

1993 64307 1714 0.0257 363 0.0059 1313 0.0201

1994 64488 1636 0.0239 444 0.0065 1272 0.0190

1995 57499 1357 0.0223 356 0.0062 1164 0.0200

1996 58939 1431 0.0226 369 0.0064 1099 0.0185

1997 58096 1466 0.0237 391 0.0066 1212 0.0204

1998 58670 1490 0.0238 426 0.0076 1178 0.0197

1999 60287 1523 0.0241 487 0.0083 1096 0.0176

2000 57956 1452 0.0235 468 0.0086 1087 0.0190

2001 95774 2436 0.0234 703 0.0078 1806 0.0184

Table 2: Number of Observations and Employment Shares, by Industry

InsuranceOther Finance IndustriesBanking


32/47


33/47


34/47


35/47


36/47


37/47

Figure 1: The Growth of the Financial Sector

Notes: Finance includes Insurance but excludes Real Estate. Value added share is the nominal value added of the financial sector divided by the nominal GDP of the U.S.

Employment includes full-time and part-time employees. Data: Annual Industrial Accounts, Bureau of Economic Analysis.

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

1947 1950 1953 1956 1959 1962 1965 1968 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004

GDP Share Compensation Share Employment Share


38/47

Notes: Share of Private Hours Worked is the number of hours worked in the sector divided by the total number of hours worked in the entire private sector. Share of Private

Labor Income is the sum of annual wages paid to employees of the financial sector divided by the sum of annual wages in the entire private sector. Hours and wages are

weighted using sampling weights. Data: March CPS.

Figure 2: Evolution of Industries

0

.01

.02

.03

1970 1980 1990 2000

Banking Other Finance

Insurance

Share of Private Hours Worked

0

.01

.02

.03

.04

1970 1980 1990 2000

Banking Other Finance

Insurance

Share of Private Labor Income


39/47

Figure 3: Relative Wages and Education in Financial Sector

Notes: Relative Wage is the percentage difference in average hourly wages between employees in the financial sector and employees in the rest of the (private) economy.

Relative Education is the share of hours worked in the financial sector accounted for by employees with at least a college degree, minus the corresponding share in the rest of

the private sector. Data: March CPS.

0

.05

.1

.15

.2

Difference

inCollegeGradu

ateShare

.1

.2

.3

.4

.5

Percent-DifferenceinHou

rlyWage

1970 1975 1980 1985 1990 1995 2000 2005

Relative Wage Relative Education


40/47

Figure 4: College Graduates and Education Intensity

Notes: Share of Hours Worked by College Graduates is the share of hours worked in the sector accounted for by employees with at least a college degree. Relative

Normalized Education Intensity is the ratio of high skilled hours over low skilled hours in the particular sector, divided by the same ratio in the rest of the private sector. The

measure is normalized so that it has a mean of one over the first five years of the sample. Hours and wages are weighted using sampling weights. Data: March CPS.

0

.2

.4

.6

.8

1970 1980 1990 2000

Banking Other FinanceInsurance Non Fin.

Share of Hours Worked by College Graduates

.8

1

1.2

1.4

1.6

1.8

1970 1980 1990 2000

Banking Other FinanceInsurance

Relative Normalized Education Intensity


41/47

Figure 5: Hour-Shares of Occupations in the Financial Sector

Notes: Hours worked by employees in a given occupation divided by total hours worked in the financial sector. Hours and wages are weighted using sampling weights. Data:

March CPS.

.2

.3

.4

.5

.6

1970 1980 1990 2000

Managers & Professionals

Clerks & Administrative

0

.02

.04

.06

.08

.1

1970 1980 1990 2000

Computers & Math.

Securities & Assets


42/47

Notes: All wages are in 2001 U.S. dollars and are weighted using sampling weights. Data: March CPS.

Figure 6: Annual Income of Engineers and Financiers

40000

50000

60000

700

00

80000

1970 1980 1990 2000

Financier Engineer

College Graduates

40000

60000

80000

100

000

120000

1970 1980 1990 2000

Financier Engineer

Post Graduates


43/47

Notes: Coefficient of Finance dummy and 95% confidence intervals from regressions of log annual wages over individual characteristics. Basic controls include race, sex,

marital status, urban residence, potential experience and its square. Education controls are dummies for education groups: 12 years of education, 13 to 15 years, 16 to 17

years, and 18 years or more. Data: March CPS.

Figure 7: Finance Wage Premium, with and without Education Controls

0

.1

.2

.3

67-70 71-75 76-80 81-85 86-90 91-95 96-00 01-05

Basic Controls Basic and Education Controls


44/47

Notes: Coefficients and 95% confidence intervals for sector dummies in regression of log annual wages over individual characteristics. Controls include race, sex, marital

status, urban residence, potential experience and its square. Education controls are dummies for education groups: 12 years of education, 13 to 15 years, 16 to 17 years, and

18 years or more. Data: March CPS.

Figure 8: Wage Premia in Finance Industries, Controlling for Education

0

.1

.2

.3

.4

.5

67-70 71-75 76-80 81-85 86-90 91-95 96-00 01-05

Banking Insurance Other Finance Industries


45/47

Notes: Coefficients of dummies interacted with education, from regressions of annual log wage on individual characteristics, Controls include race, sex, marital status, urban

residence, potential experience and its square. Data: March CPS.

Figure 9: Wage Premia in Finance, by Education Groups

.2

.4

.6

.8

1

1.2

1968 1973 1978 1983 1988 1993 1998 2003

College Graduate College Graduate in Finance

High School High School in Finance


46/47

Notes: Coefficients and 95% confidence intervals of Finance dummy in logit regression of transition from Employment to Unemployment. Controls include current log wage,

race, sex, marital status, urban residence, potential experience and its square. Education controls are dummies for education groups: 12 years of education, 13 to 15 years,

16 to 17 years, and 18 years or more. Data: March CPS.

Figure 10: Relative Unemployment Risk in Financial Sector

-.0

4

-.03

-.02

-.0

1

0

67-70 71-75 76-80 81-85 86-90 91-95 96-00 01-05


47/47

Notes: Coefficients for position in the wage distribution in logit regression of transition from employment to unemployment. Controls include current log wage, race, sex,

marital status, urban residence, potential experience and its square, as well as the position of the individual in the current income distribution: bottom third, middle third or

top third. Data: March CPS.

Figure 11: Relative Unemployment Risk in Financial Sector, sorted by Current Wage

-.04

-.02

0

67-70 71-75 76-80 81-85 86-90 91-95 96-00 01-05

0-34th Centile 34-67th Centile 67-100th Centile

philippon reshef skill biased development w13437

Documents